Feeling the Heat of High-Frequency Trading

It’s high summer here in North America, and for a lot of us, this one has been a scorcher. Media reports have been filled with coverage of heat wave after heat wave, with temperature records falling like dominoes.

But as they say, it’s not the heat, it’s the humidity, and that was painfully true in the first week of July as a slug of tropical air settled into the northeast United States. With dewpoints well into the 70s (25°C plus) and air temperatures pushing the century-mark (38°C), people suffered and systems from transportation to the electrical grid strained under the load. But as punishing as such soupy conditions are for people, there are other effects that are less well known but of critical importance to financial markets, where increased humidity can lead to billion-dollar losses for markets. Welcome to the weird world of high-frequency trading.

Time Is Money

To understand how humidity can impact something like the stock market, it pays to know a little about the mechanics of markets, and how stocks and other financial instruments are traded in the real world of the 21st century. Most people know the basics of trading stock: buy low, sell high, profit from the difference. The arrangement once was that individuals or institutions would buy a certain number of shares of a company’s stock, hold it for a while to wait for it to appreciate in value, then sell it to someone else.

While that “buy and hold” model certainly still happens, increasingly the action in the stock market has shifted from investors to speculators, and the time frame for holding the stock has dropped dramatically. Where stocks were once held for weeks or month, or perhaps even years for certain companies, increasingly sophisticated technologies have driven the hold time down to days, hours, and even seconds. The idea with these trades is still the same — buy low, sell high — but the amount the stock appreciates over short time periods is minimal, often amounting to fractions of a penny. But when it’s a fraction of a penny profit on each of a million shares, and thousands of transactions can be executed every second, the profits can really pile up.

Another critical aspect to high-frequency trading is that the traders are competing against each other to lock in their ownership of a particular stock that appears to be poised for a big purchase by another buyer. They lock in a purchase and hold onto it for the few seconds it takes for that big buy order to come in, then they dump the stock. With fast access to upcoming trades and exquisite timing, the stock will appreciate slightly during the high-frequency trader’s brief ownership and they’ll earn a small profit on each share sold.

As you can imagine, the technology needed to accomplish this is highly specialized, with custom algorithms running on server farms to gather the information needed and execute the trades. Generally such infrastructure is located close to the exchanges that act as the heart of the markets, which for the US stock market means the cluster of exchanges in New York known collectively as Wall Street. The closer the traders are to the action, the better their information will be, and the better they’ll be able to compete with other traders.

The problem is that the systems traders use need to be connected to each other, and like any other network always have to contend with latency. To minimize latency, high-frequency traders tend to locate as close of physically possible to the exchanges, as less distance to cover means less time to send and receive messages that have a speed limit determined by the laws of physics. That’s not a hard rule, of course; some traders need rapid access to transatlantic data too, and are apparently turning to HF band radio links to gain the edge over other traders.

For short-haul links around the financial centers in New York, though, dedicated network links are favored for low-latency connections. Rather than trusting their trades to the vagaries of the Internet and risk an unfavorable routing path or a cable severed by an errant backhoe, high-frequency trading firms often rely on microwave links to exchange information. The links are similar to the microwave backbone that once stitched the phone system together: point-to-point links via dedicated microwave dishes. For the Nasdaq exchange, microwave dishes link the company’s data center in Carteret, New Jersey to the New York Stock Exchange’s data center in Secaucus, New Jersey, about 16 miles to the northeast.

“Oh yeah? What exit?” A Nasdaq data center with microwave tower in Carteret, NJ, Exit 12 off the NJ Turnpike. Source: Google Maps

Sweating It Out

But what in the world does the weather have to do with all this? How can a hot, humid day possibly negatively impact the world of high-frequency trading? As it turns out, those microwave connections are the weak link in the system. During the early July heatwave, the links were experiencing slight delays in transmission times over that 16-mile path and throwing off the timing of the trading algorithms. The delay was minuscule — on the order of 10 microseconds — but in a business where millions are made and lost in seconds, that’s substantial.

The physics underlying this uniquely first-world problem are well known. We tend to think that radio waves travel at the speed of light, and while that’s true in a vacuum, propagation speed varies slightly in different media. Air makes microwaves slower; increased humidity makes them slower still. In addition, microwaves are absorbed and therefore attenuated by water vapor in the atmosphere.

So humidity deals a double whammy to the high-frequency traders’ links, both delaying the microwave signals and reducing their signal strength. Such effects are well-known, having been noticed for years on long distance telco microwave connections and the backhaul links that stitch together the cellular network. Delays and attenuation don’t really impact those services to any great degree, but even over a short path like the link between those two New Jersey data centers, the soupy air that settled into the region was enough to slow the links by a few microseconds, which is an eternity in the HFT business.

So, the next time you Northeasterners are sweltering away in your steam-bath conditions in the long, hot summer of 2018, spare a thought for those microwaves whizzing over your head with a little less zip than normal. Those signals carry a fair slice of market liquidity, and as uncomfortable as you may be in those sticky conditions, those extra microseconds are making some traders sweat for an entirely different reason.

44 thoughts on “Feeling the Heat of High-Frequency Trading

  1. And this article sums up why the stock market value of a company doesn’t say much about the companies actual value.

    And it is also interesting to see how the idea of investing in something and owning part of that investment has turned into a market where the original investment doesn’t have any real meaning. It is oddly similar to how the crypto currency market has evolved…

    1. First-world problem indeed. High frequency trading is just another sign that the market increasingly serves and enriches a smaller group of investors, and the rest of us, with our retirement contributions, are just providing feedstock.

      Crypto currencies mark the complete decoupling of investing from the idea of raising capital for businesses that support or advance society. It’s just “make something up to sell” and then try to ride it til it crashes. Even tulip bulbs had an alternate use.

        1. There exist energy efficient cryptocurrencies, including some that I’m mining with solar power. My most profitable setup uses about 13W and makes about $40/month. Giving up on cryptocurrency altogether because most are inefficient is like saying that we should give up on trains because the early ones polluted so much and were so slow.

          1. “Green” energy being used for crypto mining is energy that is not displacing traditional energy sources, so there is no valid excuse. On top of that, that energy consumed still turns into mostly hot air, which is about all crypto is. The pollution China creates producing and shipping semiconductors that we all consume is absurd enough without the added consumption. Consumer power use is only a part of the environmental impact.

          2. What crypto currency is currently giving that value?!

            If one can spend 13 watts to get 40$ a month, then I think everyone and their dog will be mining that with far more powerful equipment…
            Unless you missed the K for those watts. (But then the electricity would be worth more then 40$ a month…)

          3. “Giving up on cryptocurrency altogether because most are inefficient is like saying that we should give up on trains because the early ones polluted so much and were so slow.”

            Trains carry people and goods, and they did so right from about day one.

            I don’t doubt that blockchain will soon form part of the life-cycle of real currencies, but the idea of a stateless untraceable cyber-currency has been useless as a currency (except for crime) and has mainly served as a pseudo-asset by which some people with smarts and/or good timing have fleeced some dumb people. All while burning the energy of a small Western country.

          4. I started with Perk mining close to 3 years ago, made $100/month with 4 cheap smartphones that together use 8W or so. Some have built large mining clusters back in the day, but the main obstacle was the exchange limits placed specifically to prevent large scale mining.

            I can’t say exactly which coins I’m currently mining because I do not want to hasten the difficulty increase which would make them less efficient. I do share mining profits with those who give back to the community, so I’m not selfish. :)

            Posted by Mike Lu on Thursday, November 23, 2017

        1. But with blockchain the old hammer-on-knuckles decryption method gives the cops a nice transaction history of the ‘coin’; probably as bad or worse than using a personal credit card.

    2. Force stock to be hold for a minimum of 5 years after the buy and companies will start to plan long term, sure some parasites will go under but the market will emerge stronger after a few years

          1. It is, there are no rules or regulations stopping you doing it.

            High frequency trading makes no difference to buy and hold investors. A penny (or fraction thereof) should not make a difference to somebody with an long term perspective. Often for small retail investors they improve price knowing there is no toxicity to the orders. If you are trading so often that the tiny spread makes a difference then you should either invest in the technology to compete or get out of the game. Nobody feels sorry for me that I can’t compete with a car company out of my garage. Furthermore the entire capital base of the hft community is a fraction of that of the largest hedge funds or university endowments, look at Virtu’s annual reports if you don’t believe me. Feel free to look for a scapegoat as to why you’re unable to make money as an amateur in a professional investors world, but HFT does not have the capital to distort stock prices. Look elsewhere like infinite central bank liquidity and huge budget deficits and corporate stock buybacks, tax cuts etc. Or do what the rest of us have and invest in low cost market wide etfs which have had great returns over the last decade, two, three and so on. Actually the market has outperformed the publicly listed HFT companies. But continue to scapegoat them for your failures.

          2. So long as you can afford real estate on wall street or a microwave backbone it is.
            HFT has it’s issues but the fact that it takes money to make money isn’t unique to HFT.

        1. @Chankster: Liquidity: A measure of the extent to which a person or organization has cash to meet immediate and short-term obligations, or assets that can be quickly converted to do this.

          I doubt that Jane or John Doe would feel the need to be able to buy their groceries or sell their old car several times per second.

          I understand that HFT is good for money traders but how does it help, say RedHat?

          1. Theoretically it stabilizes the market so you at Large Bank Corp. don’t lose half of your mutual fund when another large investment firm (eg, Lehman Brothers) collapses. Like any system the rules can be gamed in unforeseen ways reducing the benefits.

      1. Some stock markets don’t allow computers to trade on them. As in algorithms, AI, scrips and so forth.
        A person can still use a computer to make the transactions, and even to analyze the market.

        Though, the general problem with HFT is that people don’t consider the value of the stock to be what the company that it is tied to is doing. The stock could literally just be a piece of paper that is tied to nothing, crypto currencies are a good example of such a market.

        Bitcoin (or other currency) can go up, or down in value, purely based on if people want/trust it or not.

        Microsoft’s (or other companies on the stock markets) stocks should though only go up or down based on if people believe or suspects that the company is going to do good or bad on the market. (Not the stock market, but the consumer and cooperate markets where they are providing their products and services.)

        HFT mostly induces noise onto the market value, and can also induce a fair bit of ringing. It is something that can make people money, but for any long term investor, it surely isn’t. (The majority of stocks don’t change hand on a 5 year period. (Yes, long term investing is common, and far more common then HFT))

        Though, the biggest downside to HFT is the fact that most stock markets takes a transaction fee, not just as a percentile, but also typically a flate rate on that as well. (And sometimes that flate rate can be rather steep.)

        Now, I haven’t traded on any New York stock market before. So I can’t speak for what they allow or not.

          1. How do one know if someone has a computer doing automated trading for them?
            Well, how do one know if someone has inside information?

            Some people make themselves obvious. (Like trading far too fast/frequently to be normal. Or making far too well informed decisions.)

            And in the case of someone actually making a decent profit on the stock markets, then investigations isn’t uncommon.

  2. HFTs already know that microwave is unreliable. They tend to have microwave and fibre links so that on a particularly rainy day when the microwave link is slow and unreliable the fibre link can take over. That doesn’t tell the whole story though since there are real time fluctuations that might make the microwave link worse than fibre at that particular moment so you can send exactly the same packets over both links and discard the copy that arrives second guaranteeing at least fibre-delay but possibly less. It’s neat!

  3. Some of the networking equipment I’ve designed has been “cut-through” (as opposed to “store-and-forward”). This means it will start forwarding a packet out the egress port as soon as it has decoded enough of the header to know what to do with the incoming packet. The traditional store-and-forward way would be to read the entire packet into a buffer memory, and only then start to send it out the egress port (after the CRC has been verified).

    The difference in latency is only some tens of ns for the smallest frames, but a whopping 1.1us or so for a 1500 byte frame (and 7-ish us for a 9k jumbo frame). (I’ve assumed a 10Gb/s connection.)

    Cut-through breaks the traditional Ethernet model in some ways. For example, an Ethernet MAC is meant to discard frames with CRC errors and the “higher” layers in the network stack never need to deal with errored frames. This isn’t possible with a cut-through system since the packet has already started to be transmitted before its end (where the CRC is) has been received.
    This in turn means that the entire packet processing datapath must be aware that the frame it’s working on right now may be corrupt, and internal state needs to be designed such that it can be rolled back in the event of an incoming CRC error. Thanks to deep pipelining, this can be quite fun.

    1. For smaller networks this method of package handling can be fairly efficient (even with corrupt packages), especially in applications where time is of the essence.

      I myself have been studying parallel computing in both multi CPU and clustered computer environments, and all forms of latency reduces what workloads can and can not be effectively executed in parallel.

      The more tightly nit workloads that actively share data on current processing tasks in real time can greatly be effected by latency. And then we haven’t even looked at other fun stuff like cache coherence and such where even nanoseconds matter…

      So I would really like if “regular” Ethernet switches were more easy to set up to just send the package forth as fast as it knows where it will go. Instead of buffering it all…. (Since tightly nit computing tasks will care about an extra microsecond or three, even 50ns makes a noticeable difference at times…)

  4. Those numbers don’t really pass the sniff test. For one, if humid air imposes an additional delay of 10 microseconds over a few km of path, then it would have a very serious impact on GPS, which depends on signal delays being known to nanoseconds.

    Over a 16-mile (26 km) path, the propogation delay is 85 microseconds in a vacuum. To add 10 microseconds would require a change in refractive index of 10/85 = 0.12. This is about 1.6% that of *pure water* (which is +7.5 at microwave frequencies). The air would need to contain 16 kg of water per cubic meter to produce that magnitude of effect.

    Air at body temperature and 100% relative humidity contains about 50 grams per cubic meter. So, to produce that 10 microsecond additional delay would need 320 times the water that 100% humidity air contains.

    The equations in “Refractive Index Formulae for Radio Waves” https://www.fig.net/resources/proceedings/fig_proceedings/fig_2002/Js28/JS28_rueger.pdf
    Suggest the *total* delay due to refraction in hot, humid air is less than 0.1%, with water accounting for half of that. Nowhere near enough to make a 85 microsecond delay into a 95 microsecond delay.

    Something else is going on, or the heat is affecting my arithmetic.

    1. I check your numbers…. Another source gives refractive index for red light at 40C 100% humidity at about 1.00025. Even doubling the difference over C for greater dispersion in microwave, and doubling the distance accounting for round trip messages the delay is barely 0.1 micro second.

      1. It’s not the dispersion (e”) that’s the issue: it’s the refractive index (e’). In the microwave region, since it’s on the low-frequency side of a monstrous resonant absorption peak around 20 GHz, water’s refractive index is around 8.5, so it’s much worse than “doubling the difference over C”, but still nowhere near enough to give the 10 us effect described.

    2. “Something else is going on” Probably.

      I don’t know what frequency they’re using, but 2.4 GHz microwave has horrible problems with water — that’s why “we” got allocated the junk band after all, and it’s why you can use 2.4 GHz to cook food.

      Do these transmitters use 2.4 GHz or thereabouts? Is it fading due to humidity? Maybe that causes them to have latency issues? Do they re-transmit on failures? Anyway, that’s an amplitude effect and not a delay.

      So yeah. Not sure what’s going on here either.

      Or is it the heat melting the towers? :)

      1. Terrestrial links use between 0.9 GHz (uncommon) through 14+ GHz. Antenna size vs gain limits you at the low end. The water absorption peak around 20 GHz limits path lengths at the high end. Oxygen gets you above that, so there’s pretty much a no-fly zone up there.

    3. I could be wrong but I think they are hinting at diffuse scattering. Consider the optical equivalent: if you shine an idealized instantaneous flash of light through a mist, a very small fraction of the light would pass through ballistically at the speed of light for that wavelength and humidity resulting in a dim but similarly instantaneous flash. The bulk of the photons are scattered (but at each straigntline section still moving at this speed of light) and thus take a wide variety of longer paths resulting in a delayed and broadened pulse of light. The delay resulting in the decreased effective speed of light, and depending on the communication system possibly also affecting its bandwidth due to broadening.

      If you look at a cloud passing before the moon, the cloud does not look like some irregular free-form lense, as it would be if we ignored diffuse scattering.

      I believe your computation to be correct for the path-lengths inbetween scattering events. Of course the scattering would depend on multiple parameters, Water seed drop size and so on.

    1. They appear to be exhibiting more and more socialist type viewpoints. Anything to knock down the bourgeosie and capitalistic success (“why do you have better things than me?” jealousy).

  5. just FYI, NYSE isn’t in Secaucus, it’s in Mahwah, NJ (MacArthur Blvd). BATS and DirectEdge markets are at 755/800 Secaucus Blvd. Lots of dark pools and banks are the Secaucus data center as well. The article does get NASDAQs location correct. Also most of the local inter-NJ links are millimeter (see Apsara) waves and not microwave links.

  6. Most of the posters here are *not* traders ! “buy low sell high” ??? yeah for the masses. Professionals use derivatives to minimize risk, maximize win probabilities. I’ve done plenty of trades where I’ve made money in either direction the market goes. Use options (a “derivative” financial instrument), combined with good money management one can make a decent mid six figure income. Sure you take losses, but that’s where money management comes in. End of the week, hopefully you show a decent ROI on the money you put at risk.

    E-Mini SP500 futures are also great instruments to trade.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.